Count each data centre once before it enters the load forecast
Rural utilities are being offered more data-centre load than they have ever served, and much of it is the same few projects knocking on several doors. A forecast should count each project once, and only after its developer has put money at risk.
I believe the most misleading number in a rural utility's planning today is the total of its data-centre enquiries. It is large, it grows every month and much of it is counted more than once. My position is that a utility should keep one register of every large-load request that it or its members receive, match each request to a parcel, a developer and a point of connection so that a project counts once, and admit a request to the base load forecast only after the developer has signed an agreement that puts its own money at risk. Everything before that point belongs in scenarios, labelled as such.
This is not an argument against data centres. Several projects in our register would be good customers for our members: large, steady loads that would spread fixed costs over more kilowatt-hours and could lower rates for farms and households. It is an argument about what a board, a lender and a regional planner are entitled to see when they are asked to approve a power station that members will pay for over thirty years.
Nine gigawatts of letters
Merrowind is a generation and transmission cooperative owned by seventeen distribution cooperatives in Oklahoma, Kansas and Arkansas. Our peak demand in 2025 was 2,640 megawatts. By early that year, the enquiries our members had passed to us from data-centre developers totalled 9.6 gigawatts, more than three times what their 412,000 meters draw on the coldest morning of the year.
The enquiries arrived as letters, emails and slide presentations. Some named a site and a substation; some named only a county; a few gave a size and a date and nothing else. Each member forwarded what it had received, and each total was added to the last. Our planners, asked by the board for a high-growth scenario, used 4 gigawatts as what seemed a cautious fraction of the whole. A proposal for a second combined-cycle plant of about 900 megawatts, estimated at US$1.6 billion, was being drafted around that scenario.
My team found the problem almost by accident. We were building a new data feed for the load forecast, which meant giving each request a record with defined fields, and the records would not stay distinct. One developer's project appeared at three members under three project names. Two developers were pursuing the same parcel. One request had been resubmitted after its developer changed consultants, and both versions had been counted. Very little of this was deceptive. Developers rationally approach several utilities at once, because a firm connection date is now the scarcest thing they buy, and each utility sees only its own correspondence.
When we matched every request by parcel, developer and nearest substation, 63 enquiries resolved to 41 distinct projects and 4.1 gigawatts. That is still a great deal of load. It was also less than half of what had been reported, and almost none of it had been committed.
Gates that cost the developer something
Removing duplicates does not tell a utility which projects will be built. For that I proposed four gates. An enquiry is a record with a site and a size. A study agreement requires a deposit towards the engineering work that tells the developer what connection would cost. A facilities agreement requires financial security for the substation and transmission work built on the developer's behalf. A signed electric service agreement fixes the load, the ramp, the term and the minimum bill. Only the last two gates enter our base forecast. Projects at the study stage appear in scenarios, and enquiries appear in a count, never in a chart of expected demand.
The gates are deliberately plain. Each asks a developer to spend money it would not spend on a project it did not intend to build, and that is the only evidence of intent a utility can verify. Letters of support from county officials, land options and announcements describe enthusiasm. Enthusiasm is not a sound basis for borrowing US$1.6 billion against members' rates.
The gates had opponents. Our vice president of economic development argued that developers would go to utilities that asked for less, and two did, taking about 700 megawatts with them. The general manager of our largest member objected that Merrowind would see his prospective customers' commercial terms. He was right to raise it, and the board added a rule I had not thought to propose: outside the planning team, our staff see a project's stage and size, never its terms. That rule has done more for members' trust in the register than anything I designed.
By June 2026, projects totalling 1.35 gigawatts had paid study deposits and 610 megawatts had signed facilities agreements. In September 2025 the board deferred the combined-cycle decision and approved two 230-megawatt combustion turbines that can be built in sequence as signed load arrives. If the deferred projects firm up, the larger plant can return to the agenda with commitments behind it.
What a register cannot settle
The limits matter. A register cannot tell a utility whether national demand from data centres will be as large as forecasters expect, and I claim no special insight into that. It does not shorten the queues in which our own generation requests have waited for years at the regional transmission organisation. Nor does it decide who should pay for transmission built for one large customer. That is a rate-design question for members' boards and, where they have jurisdiction, for regulators.
The register also has costs of its own. Matching requests takes analysts' time, and the confidentiality rules slow some conversations. Developers find the gates tiresome; one told a member manager that we were the most bureaucratic utility in three states. I accept the reputation. A cooperative that builds for load that never arrives sends the bill to farms and small towns for a generation.
One further requirement now belongs, in my view, in every signed agreement: real-time telemetry from the customer's main substation into our control centre, and contractual terms for reducing load during grid emergencies, settled before construction rather than negotiated during the first heatwave. A 300-megawatt load that the control room cannot see is an operating risk as well as a planning one.
My advice to another utility's board is narrow. At the next planning review, ask how many distinct projects sit behind the headline figure, how many have paid for a study and how many have posted security. Ask whether any project appears at more than one of your members or neighbours. If nobody can answer, the forecast describes your inbox rather than your future demand.